The Massachusetts Institute of Technology (MIT) showcased a brand new robot training model this week that abandons the previous focus on training methods centered around specific datasets, opting instead for an approach that utilizes vast amounts of information similar to that used in training large language models (LLMs). Researchers noted that imitation learning—where agents learn by mimicking individuals performing tasks—can fail when facing minor challenges. These challenges may include varying lighting conditions, different environmental setups, or new obstacles. In such cases, the robots lack sufficient data to adapt to these changes.

This week, MIT showcased a new model for training robots, aimed at addressing the challenges faced by imitation learning when introduced to small obstacles. Researchers pointed out that imitation learning can fail in conditions such as varying lighting, different environments, or new obstacles because robots simply do not have enough data to adapt. The team sought powerful data methods, like those used in models such as GPT-4, to tackle the problem. They introduced a new architecture called the Heterogeneous Pretrained Transformer (HPT), which aggregates information from different sensors and environments.

Is AI actually starting to ‘get smarter’?! The latest research from MIT indicates that the internal structure of large language models (LLMs) astonishingly resembles that of the human brain! This study utilized sparse autoencoder techniques to conduct an in-depth analysis of the activation space of LLMs, revealing three hierarchical structural features that are truly remarkable: first, at the microscopic level, researchers discovered the existence of structures akin to ‘crystals’. The surfaces of these ‘crystals’ are composed of parallelograms or trapezoids, similar to familiar lexical analogies such as ‘man:’

In August, Elon Musk's AI startup xAI promised to launch a generative AI model API named Grok, and this promise has finally come to fruition. However, the current API is relatively simple in functionality. Currently, xAI's API only offers one model, 'grok-beta', priced at $5 per million input tokens (approximately 750,000 words) or $15 per million output tokens. Tokens are essentially raw data.

Recently, OpenAI quietly released a statement on its official website, committing not to use its patents in an infringing manner. OpenAI stated that they would adhere to the principles of 'broad access' and 'collaboration', emphasizing that they would 'only use patents for defensive purposes'. However, this commitment has raised concerns among many experts. According to OpenAI's statement, their patents will be used defensively only when others do not threaten them, make claims, initiate lawsuits, or assist others in such activities. Their wording seems to suggest...

In the world of artificial intelligence, the reasoning capabilities of machine learning models, particularly large language models (LLMs), have been a focal point for scientists. Recently, Apple's AI research team published a paper titled 'Understanding the Limitations of Large Language Models in Mathematical Reasoning,' shedding light on these models' limitations when addressing logical problems. In the paper, researchers demonstrate this through a simple mathematical question. They first posed a question about Oliver picking kiwis: as follows: Oliver picked 44 on Friday.

Apple is set to launch its first AI feature - Apple Intelligence, for iPhone, iPad, and Mac devices. However, with the advancement of AI technology, privacy concerns have become increasingly important. So, how does Apple Intelligence protect user privacy? Apple's privacy protection strategy prioritizes on-device processing, followed by private cloud computing. This approach has two main advantages: first, on-device processing is faster.

On the stage of scientific research, MIT's team has just introduced a new 'Research Agent' - SciAgents, an AI system capable of conducting research autonomously. Its powerful capabilities are truly impressive. In research on bio-inspired materials, SciAgents revealed some previously considered unrelated interdisciplinary connections, successfully achieving a scale, precision, and exploratory capacity that surpass conventional human research. SciAgents is a 'super assistant' in the research community. This intelligent system can independently read literature, confirm

Tencent Zhi Ying has launched a new feature called 'AI Online Sensation Helper', aimed at simplifying the photo editing process and enhancing efficiency through artificial intelligence technology. This feature, using a cloud-based smart video creation tool, allows users to upload photos and have them automatically beautified and optimized by AI, including portrait beautification and background enhancement, to meet users' pursuit of trendiness and personalization.

Recently, a study from the Swiss École Polytechnique Fédérale de Lausanne (EPFL) made its debut at the IEEE International Conference on Robotics and Automation in Rotterdam. This research aims to explore how robotic hands can break existing limitations to grasp more objects. The research team noted that deep learning models significantly enhanced the dexterous manipulation capabilities of multi-fingered hands, but the grasping guided by contact information in cluttered environments has not yet been fully explored. To address this issue, the researchers designed an inspired hand that can bend backward to pick up various objects and detach itself to climb.

Recently, scientists at the Polytechnic University of Valencia in Spain conducted a study revealing that large language models such as GPT, LLaMA, and BLOOM tend to lie rather than admit ignorance when faced with questions. The research found that as the complexity of AI models increases, their accuracy in tackling complex issues decreases, and they are more likely to fabricate answers. Researchers noted that human volunteers also struggled to identify these incorrect answers during tests, suggesting a potential risk posed by AI lies to humans. Scientists recommend improving AI.

Shanghai Mita Network Technology Co., Ltd. announced the launch of a new image search feature on its Mita AI Search platform. The new feature allows users to view related images categorized, making the search process more organized and saving users' time and effort. Additionally, the feature provides users with recommendations for scenic spots and routes, helping them better plan their travels.

Recently, U.S. officials announced that a global Artificial Intelligence (AI) Safety Summit will be hosted in San Francisco in November. As AI technology advances rapidly, governments around the world are striving to keep pace with this change. The summit will be co-chaired by U.S. Secretary of Commerce Gina Raimondo and Secretary of State Antony Blinken, and is scheduled for November 20-21. Raimondo stated in a release, 'As AI develops rapidly, we are doing everything we can.' She noted the importance of close international collaboration and thoughtful coordination.

Recently, OpenAI's CEO Sam Altman announced his decision to leave the company's internal safety committee, which was established in May of this year to oversee key safety and security decisions in the development and deployment of OpenAI's AI models. Altman's decision has drawn significant attention as he hopes to hand this role over to third-party experts to enhance transparency, collaboration, and security. With Altman's departure, this safety committee will transform into an 'independent oversight committee' focused on safety and security issues.